Schema evolution is a crucial aspect in database management. The proposed taxonomies of schema changes have neglected the set of operations that involves relationships between entity types: aggregation and references, as well as the possible existence of structural variations for schema types, as most of NoSQL systems are schemaless. The distinction between entity types and relationship types, which is typical of graph schemas, is also not taken into account in the published works. Moreover, NoSQL schema evolution poses the challenge of having different data models, and no standard specification exists for them. In this paper, a generic approach for evolving NoSQL and relational schemas is presented, which is based on the U-Schema unified data model that includes aggregation and reference relationships, and structural variations. For this data model, we introduce a taxonomy of schema changes for all the U-Schema elements, which is implemented by creating the Orion database-independent language. We will show how Orion can be used to automatically generate evolution scripts for a set of NoSQL databases, and the feasibility of each schema operation will be analyzed through the performance results obtained. The taxonomy has been formally validated by means of Alloy, and two case studies show the application of Orion.